2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *queue)
214 spin_lock(&queue->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *queue)
221 spin_unlock(&queue->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
960 if (!dev_valid_name(name))
963 if (fmt && strchr(name, '%'))
964 return __dev_alloc_name(net, name, buf);
965 else if (__dev_get_by_name(net, name))
967 else if (buf != name)
968 strlcpy(buf, name, IFNAMSIZ);
974 * dev_change_name - change name of a device
976 * @newname: name (or format string) must be at least IFNAMSIZ
978 * Change name of a device, can pass format strings "eth%d".
981 int dev_change_name(struct net_device *dev, const char *newname)
983 char oldname[IFNAMSIZ];
989 BUG_ON(!dev_net(dev));
992 if (dev->flags & IFF_UP)
995 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
998 memcpy(oldname, dev->name, IFNAMSIZ);
1000 err = dev_get_valid_name(net, newname, dev->name, 1);
1005 /* For now only devices in the initial network namespace
1008 if (net_eq(net, &init_net)) {
1009 ret = device_rename(&dev->dev, dev->name);
1011 memcpy(dev->name, oldname, IFNAMSIZ);
1016 write_lock_bh(&dev_base_lock);
1017 hlist_del(&dev->name_hlist);
1018 write_unlock_bh(&dev_base_lock);
1022 write_lock_bh(&dev_base_lock);
1023 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1024 write_unlock_bh(&dev_base_lock);
1026 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1027 ret = notifier_to_errno(ret);
1030 /* err >= 0 after dev_alloc_name() or stores the first errno */
1033 memcpy(dev->name, oldname, IFNAMSIZ);
1037 "%s: name change rollback failed: %d.\n",
1046 * dev_set_alias - change ifalias of a device
1048 * @alias: name up to IFALIASZ
1049 * @len: limit of bytes to copy from info
1051 * Set ifalias for a device,
1053 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1057 if (len >= IFALIASZ)
1062 kfree(dev->ifalias);
1063 dev->ifalias = NULL;
1068 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1072 strlcpy(dev->ifalias, alias, len+1);
1078 * netdev_features_change - device changes features
1079 * @dev: device to cause notification
1081 * Called to indicate a device has changed features.
1083 void netdev_features_change(struct net_device *dev)
1085 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1087 EXPORT_SYMBOL(netdev_features_change);
1090 * netdev_state_change - device changes state
1091 * @dev: device to cause notification
1093 * Called to indicate a device has changed state. This function calls
1094 * the notifier chains for netdev_chain and sends a NEWLINK message
1095 * to the routing socket.
1097 void netdev_state_change(struct net_device *dev)
1099 if (dev->flags & IFF_UP) {
1100 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1101 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1104 EXPORT_SYMBOL(netdev_state_change);
1106 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1108 return call_netdevice_notifiers(event, dev);
1110 EXPORT_SYMBOL(netdev_bonding_change);
1113 * dev_load - load a network module
1114 * @net: the applicable net namespace
1115 * @name: name of interface
1117 * If a network interface is not present and the process has suitable
1118 * privileges this function loads the module. If module loading is not
1119 * available in this kernel then it becomes a nop.
1122 void dev_load(struct net *net, const char *name)
1124 struct net_device *dev;
1127 dev = dev_get_by_name_rcu(net, name);
1130 if (!dev && capable(CAP_NET_ADMIN))
1131 request_module("%s", name);
1133 EXPORT_SYMBOL(dev_load);
1135 static int __dev_open(struct net_device *dev)
1137 const struct net_device_ops *ops = dev->netdev_ops;
1143 * Is it even present?
1145 if (!netif_device_present(dev))
1148 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1149 ret = notifier_to_errno(ret);
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START, &dev->state);
1158 if (ops->ndo_validate_addr)
1159 ret = ops->ndo_validate_addr(dev);
1161 if (!ret && ops->ndo_open)
1162 ret = ops->ndo_open(dev);
1165 * If it went open OK then:
1169 clear_bit(__LINK_STATE_START, &dev->state);
1174 dev->flags |= IFF_UP;
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev);
1187 * Wakeup transmit queue engine
1196 * dev_open - prepare an interface for use.
1197 * @dev: device to open
1199 * Takes a device from down to up state. The device's private open
1200 * function is invoked and then the multicast lists are loaded. Finally
1201 * the device is moved into the up state and a %NETDEV_UP message is
1202 * sent to the netdev notifier chain.
1204 * Calling this function on an active interface is a nop. On a failure
1205 * a negative errno code is returned.
1207 int dev_open(struct net_device *dev)
1214 if (dev->flags & IFF_UP)
1220 ret = __dev_open(dev);
1225 * ... and announce new interface.
1227 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1228 call_netdevice_notifiers(NETDEV_UP, dev);
1232 EXPORT_SYMBOL(dev_open);
1234 static int __dev_close(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1242 * Tell people we are going down, so that they can
1243 * prepare to death, when device is still operating.
1245 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1247 clear_bit(__LINK_STATE_START, &dev->state);
1249 /* Synchronize to scheduled poll. We cannot touch poll list,
1250 * it can be even on different cpu. So just clear netif_running().
1252 * dev->stop() will invoke napi_disable() on all of it's
1253 * napi_struct instances on this device.
1255 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1257 dev_deactivate(dev);
1260 * Call the device specific close. This cannot fail.
1261 * Only if device is UP
1263 * We allow it to be called even after a DETACH hot-plug
1270 * Device is now down.
1273 dev->flags &= ~IFF_UP;
1278 net_dmaengine_put();
1284 * dev_close - shutdown an interface.
1285 * @dev: device to shutdown
1287 * This function moves an active device into down state. A
1288 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1289 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1292 int dev_close(struct net_device *dev)
1294 if (!(dev->flags & IFF_UP))
1300 * Tell people we are down
1302 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1303 call_netdevice_notifiers(NETDEV_DOWN, dev);
1307 EXPORT_SYMBOL(dev_close);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device *dev)
1320 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1321 dev->ethtool_ops->set_flags) {
1322 u32 flags = dev->ethtool_ops->get_flags(dev);
1323 if (flags & ETH_FLAG_LRO) {
1324 flags &= ~ETH_FLAG_LRO;
1325 dev->ethtool_ops->set_flags(dev, flags);
1328 WARN_ON(dev->features & NETIF_F_LRO);
1330 EXPORT_SYMBOL(dev_disable_lro);
1333 static int dev_boot_phase = 1;
1336 * Device change register/unregister. These are not inline or static
1337 * as we export them to the world.
1341 * register_netdevice_notifier - register a network notifier block
1344 * Register a notifier to be called when network device events occur.
1345 * The notifier passed is linked into the kernel structures and must
1346 * not be reused until it has been unregistered. A negative errno code
1347 * is returned on a failure.
1349 * When registered all registration and up events are replayed
1350 * to the new notifier to allow device to have a race free
1351 * view of the network device list.
1354 int register_netdevice_notifier(struct notifier_block *nb)
1356 struct net_device *dev;
1357 struct net_device *last;
1362 err = raw_notifier_chain_register(&netdev_chain, nb);
1368 for_each_netdev(net, dev) {
1369 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1370 err = notifier_to_errno(err);
1374 if (!(dev->flags & IFF_UP))
1377 nb->notifier_call(nb, NETDEV_UP, dev);
1388 for_each_netdev(net, dev) {
1392 if (dev->flags & IFF_UP) {
1393 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1397 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1401 raw_notifier_chain_unregister(&netdev_chain, nb);
1404 EXPORT_SYMBOL(register_netdevice_notifier);
1407 * unregister_netdevice_notifier - unregister a network notifier block
1410 * Unregister a notifier previously registered by
1411 * register_netdevice_notifier(). The notifier is unlinked into the
1412 * kernel structures and may then be reused. A negative errno code
1413 * is returned on a failure.
1416 int unregister_netdevice_notifier(struct notifier_block *nb)
1421 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1425 EXPORT_SYMBOL(unregister_netdevice_notifier);
1428 * call_netdevice_notifiers - call all network notifier blocks
1429 * @val: value passed unmodified to notifier function
1430 * @dev: net_device pointer passed unmodified to notifier function
1432 * Call all network notifier blocks. Parameters and return value
1433 * are as for raw_notifier_call_chain().
1436 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1438 return raw_notifier_call_chain(&netdev_chain, val, dev);
1441 /* When > 0 there are consumers of rx skb time stamps */
1442 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1444 void net_enable_timestamp(void)
1446 atomic_inc(&netstamp_needed);
1448 EXPORT_SYMBOL(net_enable_timestamp);
1450 void net_disable_timestamp(void)
1452 atomic_dec(&netstamp_needed);
1454 EXPORT_SYMBOL(net_disable_timestamp);
1456 static inline void net_timestamp(struct sk_buff *skb)
1458 if (atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1461 skb->tstamp.tv64 = 0;
1465 * dev_forward_skb - loopback an skb to another netif
1467 * @dev: destination network device
1468 * @skb: buffer to forward
1471 * NET_RX_SUCCESS (no congestion)
1472 * NET_RX_DROP (packet was dropped)
1474 * dev_forward_skb can be used for injecting an skb from the
1475 * start_xmit function of one device into the receive queue
1476 * of another device.
1478 * The receiving device may be in another namespace, so
1479 * we have to clear all information in the skb that could
1480 * impact namespace isolation.
1482 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1486 if (!(dev->flags & IFF_UP))
1489 if (skb->len > (dev->mtu + dev->hard_header_len))
1492 skb_set_dev(skb, dev);
1493 skb->tstamp.tv64 = 0;
1494 skb->pkt_type = PACKET_HOST;
1495 skb->protocol = eth_type_trans(skb, dev);
1496 return netif_rx(skb);
1498 EXPORT_SYMBOL_GPL(dev_forward_skb);
1501 * Support routine. Sends outgoing frames to any network
1502 * taps currently in use.
1505 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1507 struct packet_type *ptype;
1509 #ifdef CONFIG_NET_CLS_ACT
1510 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1517 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1518 /* Never send packets back to the socket
1519 * they originated from - MvS (miquels@drinkel.ow.org)
1521 if ((ptype->dev == dev || !ptype->dev) &&
1522 (ptype->af_packet_priv == NULL ||
1523 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1524 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1528 /* skb->nh should be correctly
1529 set by sender, so that the second statement is
1530 just protection against buggy protocols.
1532 skb_reset_mac_header(skb2);
1534 if (skb_network_header(skb2) < skb2->data ||
1535 skb2->network_header > skb2->tail) {
1536 if (net_ratelimit())
1537 printk(KERN_CRIT "protocol %04x is "
1539 skb2->protocol, dev->name);
1540 skb_reset_network_header(skb2);
1543 skb2->transport_header = skb2->network_header;
1544 skb2->pkt_type = PACKET_OUTGOING;
1545 ptype->func(skb2, skb->dev, ptype, skb->dev);
1552 static inline void __netif_reschedule(struct Qdisc *q)
1554 struct softnet_data *sd;
1555 unsigned long flags;
1557 local_irq_save(flags);
1558 sd = &__get_cpu_var(softnet_data);
1559 q->next_sched = sd->output_queue;
1560 sd->output_queue = q;
1561 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1562 local_irq_restore(flags);
1565 void __netif_schedule(struct Qdisc *q)
1567 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1568 __netif_reschedule(q);
1570 EXPORT_SYMBOL(__netif_schedule);
1572 void dev_kfree_skb_irq(struct sk_buff *skb)
1574 if (atomic_dec_and_test(&skb->users)) {
1575 struct softnet_data *sd;
1576 unsigned long flags;
1578 local_irq_save(flags);
1579 sd = &__get_cpu_var(softnet_data);
1580 skb->next = sd->completion_queue;
1581 sd->completion_queue = skb;
1582 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1583 local_irq_restore(flags);
1586 EXPORT_SYMBOL(dev_kfree_skb_irq);
1588 void dev_kfree_skb_any(struct sk_buff *skb)
1590 if (in_irq() || irqs_disabled())
1591 dev_kfree_skb_irq(skb);
1595 EXPORT_SYMBOL(dev_kfree_skb_any);
1599 * netif_device_detach - mark device as removed
1600 * @dev: network device
1602 * Mark device as removed from system and therefore no longer available.
1604 void netif_device_detach(struct net_device *dev)
1606 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1607 netif_running(dev)) {
1608 netif_tx_stop_all_queues(dev);
1611 EXPORT_SYMBOL(netif_device_detach);
1614 * netif_device_attach - mark device as attached
1615 * @dev: network device
1617 * Mark device as attached from system and restart if needed.
1619 void netif_device_attach(struct net_device *dev)
1621 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1622 netif_running(dev)) {
1623 netif_tx_wake_all_queues(dev);
1624 __netdev_watchdog_up(dev);
1627 EXPORT_SYMBOL(netif_device_attach);
1629 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1631 return ((features & NETIF_F_GEN_CSUM) ||
1632 ((features & NETIF_F_IP_CSUM) &&
1633 protocol == htons(ETH_P_IP)) ||
1634 ((features & NETIF_F_IPV6_CSUM) &&
1635 protocol == htons(ETH_P_IPV6)) ||
1636 ((features & NETIF_F_FCOE_CRC) &&
1637 protocol == htons(ETH_P_FCOE)));
1640 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1642 if (can_checksum_protocol(dev->features, skb->protocol))
1645 if (skb->protocol == htons(ETH_P_8021Q)) {
1646 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1647 if (can_checksum_protocol(dev->features & dev->vlan_features,
1648 veh->h_vlan_encapsulated_proto))
1656 * skb_dev_set -- assign a new device to a buffer
1657 * @skb: buffer for the new device
1658 * @dev: network device
1660 * If an skb is owned by a device already, we have to reset
1661 * all data private to the namespace a device belongs to
1662 * before assigning it a new device.
1664 #ifdef CONFIG_NET_NS
1665 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1668 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1671 skb_init_secmark(skb);
1675 skb->ipvs_property = 0;
1676 #ifdef CONFIG_NET_SCHED
1682 EXPORT_SYMBOL(skb_set_dev);
1683 #endif /* CONFIG_NET_NS */
1686 * Invalidate hardware checksum when packet is to be mangled, and
1687 * complete checksum manually on outgoing path.
1689 int skb_checksum_help(struct sk_buff *skb)
1692 int ret = 0, offset;
1694 if (skb->ip_summed == CHECKSUM_COMPLETE)
1695 goto out_set_summed;
1697 if (unlikely(skb_shinfo(skb)->gso_size)) {
1698 /* Let GSO fix up the checksum. */
1699 goto out_set_summed;
1702 offset = skb->csum_start - skb_headroom(skb);
1703 BUG_ON(offset >= skb_headlen(skb));
1704 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1706 offset += skb->csum_offset;
1707 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1709 if (skb_cloned(skb) &&
1710 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1711 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1716 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1718 skb->ip_summed = CHECKSUM_NONE;
1722 EXPORT_SYMBOL(skb_checksum_help);
1725 * skb_gso_segment - Perform segmentation on skb.
1726 * @skb: buffer to segment
1727 * @features: features for the output path (see dev->features)
1729 * This function segments the given skb and returns a list of segments.
1731 * It may return NULL if the skb requires no segmentation. This is
1732 * only possible when GSO is used for verifying header integrity.
1734 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1736 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1737 struct packet_type *ptype;
1738 __be16 type = skb->protocol;
1741 skb_reset_mac_header(skb);
1742 skb->mac_len = skb->network_header - skb->mac_header;
1743 __skb_pull(skb, skb->mac_len);
1745 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1746 struct net_device *dev = skb->dev;
1747 struct ethtool_drvinfo info = {};
1749 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1750 dev->ethtool_ops->get_drvinfo(dev, &info);
1752 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1754 info.driver, dev ? dev->features : 0L,
1755 skb->sk ? skb->sk->sk_route_caps : 0L,
1756 skb->len, skb->data_len, skb->ip_summed);
1758 if (skb_header_cloned(skb) &&
1759 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1760 return ERR_PTR(err);
1764 list_for_each_entry_rcu(ptype,
1765 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1766 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1767 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1768 err = ptype->gso_send_check(skb);
1769 segs = ERR_PTR(err);
1770 if (err || skb_gso_ok(skb, features))
1772 __skb_push(skb, (skb->data -
1773 skb_network_header(skb)));
1775 segs = ptype->gso_segment(skb, features);
1781 __skb_push(skb, skb->data - skb_mac_header(skb));
1785 EXPORT_SYMBOL(skb_gso_segment);
1787 /* Take action when hardware reception checksum errors are detected. */
1789 void netdev_rx_csum_fault(struct net_device *dev)
1791 if (net_ratelimit()) {
1792 printk(KERN_ERR "%s: hw csum failure.\n",
1793 dev ? dev->name : "<unknown>");
1797 EXPORT_SYMBOL(netdev_rx_csum_fault);
1800 /* Actually, we should eliminate this check as soon as we know, that:
1801 * 1. IOMMU is present and allows to map all the memory.
1802 * 2. No high memory really exists on this machine.
1805 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1807 #ifdef CONFIG_HIGHMEM
1809 if (!(dev->features & NETIF_F_HIGHDMA)) {
1810 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1811 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1815 if (PCI_DMA_BUS_IS_PHYS) {
1816 struct device *pdev = dev->dev.parent;
1820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1821 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1822 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1831 void (*destructor)(struct sk_buff *skb);
1834 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1836 static void dev_gso_skb_destructor(struct sk_buff *skb)
1838 struct dev_gso_cb *cb;
1841 struct sk_buff *nskb = skb->next;
1843 skb->next = nskb->next;
1846 } while (skb->next);
1848 cb = DEV_GSO_CB(skb);
1850 cb->destructor(skb);
1854 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1855 * @skb: buffer to segment
1857 * This function segments the given skb and stores the list of segments
1860 static int dev_gso_segment(struct sk_buff *skb)
1862 struct net_device *dev = skb->dev;
1863 struct sk_buff *segs;
1864 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1867 segs = skb_gso_segment(skb, features);
1869 /* Verifying header integrity only. */
1874 return PTR_ERR(segs);
1877 DEV_GSO_CB(skb)->destructor = skb->destructor;
1878 skb->destructor = dev_gso_skb_destructor;
1883 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1884 struct netdev_queue *txq)
1886 const struct net_device_ops *ops = dev->netdev_ops;
1887 int rc = NETDEV_TX_OK;
1889 if (likely(!skb->next)) {
1890 if (!list_empty(&ptype_all))
1891 dev_queue_xmit_nit(skb, dev);
1893 if (netif_needs_gso(dev, skb)) {
1894 if (unlikely(dev_gso_segment(skb)))
1901 * If device doesnt need skb->dst, release it right now while
1902 * its hot in this cpu cache
1904 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1907 rc = ops->ndo_start_xmit(skb, dev);
1908 if (rc == NETDEV_TX_OK)
1909 txq_trans_update(txq);
1911 * TODO: if skb_orphan() was called by
1912 * dev->hard_start_xmit() (for example, the unmodified
1913 * igb driver does that; bnx2 doesn't), then
1914 * skb_tx_software_timestamp() will be unable to send
1915 * back the time stamp.
1917 * How can this be prevented? Always create another
1918 * reference to the socket before calling
1919 * dev->hard_start_xmit()? Prevent that skb_orphan()
1920 * does anything in dev->hard_start_xmit() by clearing
1921 * the skb destructor before the call and restoring it
1922 * afterwards, then doing the skb_orphan() ourselves?
1929 struct sk_buff *nskb = skb->next;
1931 skb->next = nskb->next;
1935 * If device doesnt need nskb->dst, release it right now while
1936 * its hot in this cpu cache
1938 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1941 rc = ops->ndo_start_xmit(nskb, dev);
1942 if (unlikely(rc != NETDEV_TX_OK)) {
1943 if (rc & ~NETDEV_TX_MASK)
1944 goto out_kfree_gso_skb;
1945 nskb->next = skb->next;
1949 txq_trans_update(txq);
1950 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1951 return NETDEV_TX_BUSY;
1952 } while (skb->next);
1955 if (likely(skb->next == NULL))
1956 skb->destructor = DEV_GSO_CB(skb)->destructor;
1962 static u32 hashrnd __read_mostly;
1964 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1968 if (skb_rx_queue_recorded(skb)) {
1969 hash = skb_get_rx_queue(skb);
1970 while (unlikely(hash >= dev->real_num_tx_queues))
1971 hash -= dev->real_num_tx_queues;
1975 if (skb->sk && skb->sk->sk_hash)
1976 hash = skb->sk->sk_hash;
1978 hash = skb->protocol;
1980 hash = jhash_1word(hash, hashrnd);
1982 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1984 EXPORT_SYMBOL(skb_tx_hash);
1986 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1988 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1989 if (net_ratelimit()) {
1990 pr_warning("%s selects TX queue %d, but "
1991 "real number of TX queues is %d\n",
1992 dev->name, queue_index, dev->real_num_tx_queues);
1999 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2000 struct sk_buff *skb)
2003 struct sock *sk = skb->sk;
2005 if (sk_tx_queue_recorded(sk)) {
2006 queue_index = sk_tx_queue_get(sk);
2008 const struct net_device_ops *ops = dev->netdev_ops;
2010 if (ops->ndo_select_queue) {
2011 queue_index = ops->ndo_select_queue(dev, skb);
2012 queue_index = dev_cap_txqueue(dev, queue_index);
2015 if (dev->real_num_tx_queues > 1)
2016 queue_index = skb_tx_hash(dev, skb);
2018 if (sk && rcu_dereference_check(sk->sk_dst_cache, 1))
2019 sk_tx_queue_set(sk, queue_index);
2023 skb_set_queue_mapping(skb, queue_index);
2024 return netdev_get_tx_queue(dev, queue_index);
2027 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2028 struct net_device *dev,
2029 struct netdev_queue *txq)
2031 spinlock_t *root_lock = qdisc_lock(q);
2034 spin_lock(root_lock);
2035 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2038 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2039 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2041 * This is a work-conserving queue; there are no old skbs
2042 * waiting to be sent out; and the qdisc is not running -
2043 * xmit the skb directly.
2045 __qdisc_update_bstats(q, skb->len);
2046 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2049 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2051 rc = NET_XMIT_SUCCESS;
2053 rc = qdisc_enqueue_root(skb, q);
2056 spin_unlock(root_lock);
2062 * Returns true if either:
2063 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2064 * 2. skb is fragmented and the device does not support SG, or if
2065 * at least one of fragments is in highmem and device does not
2066 * support DMA from it.
2068 static inline int skb_needs_linearize(struct sk_buff *skb,
2069 struct net_device *dev)
2071 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2072 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2073 illegal_highdma(dev, skb)));
2077 * dev_queue_xmit - transmit a buffer
2078 * @skb: buffer to transmit
2080 * Queue a buffer for transmission to a network device. The caller must
2081 * have set the device and priority and built the buffer before calling
2082 * this function. The function can be called from an interrupt.
2084 * A negative errno code is returned on a failure. A success does not
2085 * guarantee the frame will be transmitted as it may be dropped due
2086 * to congestion or traffic shaping.
2088 * -----------------------------------------------------------------------------------
2089 * I notice this method can also return errors from the queue disciplines,
2090 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2093 * Regardless of the return value, the skb is consumed, so it is currently
2094 * difficult to retry a send to this method. (You can bump the ref count
2095 * before sending to hold a reference for retry if you are careful.)
2097 * When calling this method, interrupts MUST be enabled. This is because
2098 * the BH enable code must have IRQs enabled so that it will not deadlock.
2101 int dev_queue_xmit(struct sk_buff *skb)
2103 struct net_device *dev = skb->dev;
2104 struct netdev_queue *txq;
2108 /* GSO will handle the following emulations directly. */
2109 if (netif_needs_gso(dev, skb))
2112 /* Convert a paged skb to linear, if required */
2113 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2116 /* If packet is not checksummed and device does not support
2117 * checksumming for this protocol, complete checksumming here.
2119 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2120 skb_set_transport_header(skb, skb->csum_start -
2122 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2127 /* Disable soft irqs for various locks below. Also
2128 * stops preemption for RCU.
2132 txq = dev_pick_tx(dev, skb);
2133 q = rcu_dereference_bh(txq->qdisc);
2135 #ifdef CONFIG_NET_CLS_ACT
2136 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2139 rc = __dev_xmit_skb(skb, q, dev, txq);
2143 /* The device has no queue. Common case for software devices:
2144 loopback, all the sorts of tunnels...
2146 Really, it is unlikely that netif_tx_lock protection is necessary
2147 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2149 However, it is possible, that they rely on protection
2152 Check this and shot the lock. It is not prone from deadlocks.
2153 Either shot noqueue qdisc, it is even simpler 8)
2155 if (dev->flags & IFF_UP) {
2156 int cpu = smp_processor_id(); /* ok because BHs are off */
2158 if (txq->xmit_lock_owner != cpu) {
2160 HARD_TX_LOCK(dev, txq, cpu);
2162 if (!netif_tx_queue_stopped(txq)) {
2163 rc = dev_hard_start_xmit(skb, dev, txq);
2164 if (dev_xmit_complete(rc)) {
2165 HARD_TX_UNLOCK(dev, txq);
2169 HARD_TX_UNLOCK(dev, txq);
2170 if (net_ratelimit())
2171 printk(KERN_CRIT "Virtual device %s asks to "
2172 "queue packet!\n", dev->name);
2174 /* Recursion is detected! It is possible,
2176 if (net_ratelimit())
2177 printk(KERN_CRIT "Dead loop on virtual device "
2178 "%s, fix it urgently!\n", dev->name);
2183 rcu_read_unlock_bh();
2189 rcu_read_unlock_bh();
2192 EXPORT_SYMBOL(dev_queue_xmit);
2195 /*=======================================================================
2197 =======================================================================*/
2199 int netdev_max_backlog __read_mostly = 1000;
2200 int netdev_budget __read_mostly = 300;
2201 int weight_p __read_mostly = 64; /* old backlog weight */
2203 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2207 * get_rps_cpu is called from netif_receive_skb and returns the target
2208 * CPU from the RPS map of the receiving queue for a given skb.
2210 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb)
2212 struct ipv6hdr *ip6;
2214 struct netdev_rx_queue *rxqueue;
2215 struct rps_map *map;
2218 u32 addr1, addr2, ports, ihl;
2222 if (skb_rx_queue_recorded(skb)) {
2223 u16 index = skb_get_rx_queue(skb);
2224 if (unlikely(index >= dev->num_rx_queues)) {
2225 if (net_ratelimit()) {
2226 pr_warning("%s received packet on queue "
2227 "%u, but number of RX queues is %u\n",
2228 dev->name, index, dev->num_rx_queues);
2232 rxqueue = dev->_rx + index;
2236 if (!rxqueue->rps_map)
2240 goto got_hash; /* Skip hash computation on packet header */
2242 switch (skb->protocol) {
2243 case __constant_htons(ETH_P_IP):
2244 if (!pskb_may_pull(skb, sizeof(*ip)))
2247 ip = (struct iphdr *) skb->data;
2248 ip_proto = ip->protocol;
2253 case __constant_htons(ETH_P_IPV6):
2254 if (!pskb_may_pull(skb, sizeof(*ip6)))
2257 ip6 = (struct ipv6hdr *) skb->data;
2258 ip_proto = ip6->nexthdr;
2259 addr1 = ip6->saddr.s6_addr32[3];
2260 addr2 = ip6->daddr.s6_addr32[3];
2274 case IPPROTO_UDPLITE:
2275 if (pskb_may_pull(skb, (ihl * 4) + 4))
2276 ports = *((u32 *) (skb->data + (ihl * 4)));
2283 skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd);
2288 map = rcu_dereference(rxqueue->rps_map);
2290 u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2292 if (cpu_online(tcpu)) {
2304 * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled
2305 * to be sent to kick remote softirq processing. There are two masks since
2306 * the sending of IPIs must be done with interrupts enabled. The select field
2307 * indicates the current mask that enqueue_backlog uses to schedule IPIs.
2308 * select is flipped before net_rps_action is called while still under lock,
2309 * net_rps_action then uses the non-selected mask to send the IPIs and clears
2310 * it without conflicting with enqueue_backlog operation.
2312 struct rps_remote_softirq_cpus {
2316 static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus);
2318 /* Called from hardirq (IPI) context */
2319 static void trigger_softirq(void *data)
2321 struct softnet_data *queue = data;
2322 __napi_schedule(&queue->backlog);
2323 __get_cpu_var(netdev_rx_stat).received_rps++;
2325 #endif /* CONFIG_SMP */
2328 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2329 * queue (may be a remote CPU queue).
2331 static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
2333 struct softnet_data *queue;
2334 unsigned long flags;
2336 queue = &per_cpu(softnet_data, cpu);
2338 local_irq_save(flags);
2339 __get_cpu_var(netdev_rx_stat).total++;
2342 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2343 if (queue->input_pkt_queue.qlen) {
2345 __skb_queue_tail(&queue->input_pkt_queue, skb);
2347 local_irq_restore(flags);
2348 return NET_RX_SUCCESS;
2351 /* Schedule NAPI for backlog device */
2352 if (napi_schedule_prep(&queue->backlog)) {
2354 if (cpu != smp_processor_id()) {
2355 struct rps_remote_softirq_cpus *rcpus =
2356 &__get_cpu_var(rps_remote_softirq_cpus);
2358 cpu_set(cpu, rcpus->mask[rcpus->select]);
2359 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2361 __napi_schedule(&queue->backlog);
2363 __napi_schedule(&queue->backlog);
2371 __get_cpu_var(netdev_rx_stat).dropped++;
2372 local_irq_restore(flags);
2379 * netif_rx - post buffer to the network code
2380 * @skb: buffer to post
2382 * This function receives a packet from a device driver and queues it for
2383 * the upper (protocol) levels to process. It always succeeds. The buffer
2384 * may be dropped during processing for congestion control or by the
2388 * NET_RX_SUCCESS (no congestion)
2389 * NET_RX_DROP (packet was dropped)
2393 int netif_rx(struct sk_buff *skb)
2397 /* if netpoll wants it, pretend we never saw it */
2398 if (netpoll_rx(skb))
2401 if (!skb->tstamp.tv64)
2405 cpu = get_rps_cpu(skb->dev, skb);
2407 cpu = smp_processor_id();
2409 cpu = smp_processor_id();
2412 return enqueue_to_backlog(skb, cpu);
2414 EXPORT_SYMBOL(netif_rx);
2416 int netif_rx_ni(struct sk_buff *skb)
2421 err = netif_rx(skb);
2422 if (local_softirq_pending())
2428 EXPORT_SYMBOL(netif_rx_ni);
2430 static void net_tx_action(struct softirq_action *h)
2432 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2434 if (sd->completion_queue) {
2435 struct sk_buff *clist;
2437 local_irq_disable();
2438 clist = sd->completion_queue;
2439 sd->completion_queue = NULL;
2443 struct sk_buff *skb = clist;
2444 clist = clist->next;
2446 WARN_ON(atomic_read(&skb->users));
2451 if (sd->output_queue) {
2454 local_irq_disable();
2455 head = sd->output_queue;
2456 sd->output_queue = NULL;
2460 struct Qdisc *q = head;
2461 spinlock_t *root_lock;
2463 head = head->next_sched;
2465 root_lock = qdisc_lock(q);
2466 if (spin_trylock(root_lock)) {
2467 smp_mb__before_clear_bit();
2468 clear_bit(__QDISC_STATE_SCHED,
2471 spin_unlock(root_lock);
2473 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2475 __netif_reschedule(q);
2477 smp_mb__before_clear_bit();
2478 clear_bit(__QDISC_STATE_SCHED,
2486 static inline int deliver_skb(struct sk_buff *skb,
2487 struct packet_type *pt_prev,
2488 struct net_device *orig_dev)
2490 atomic_inc(&skb->users);
2491 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2494 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2496 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2497 /* This hook is defined here for ATM LANE */
2498 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2499 unsigned char *addr) __read_mostly;
2500 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2504 * If bridge module is loaded call bridging hook.
2505 * returns NULL if packet was consumed.
2507 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2508 struct sk_buff *skb) __read_mostly;
2509 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2511 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2512 struct packet_type **pt_prev, int *ret,
2513 struct net_device *orig_dev)
2515 struct net_bridge_port *port;
2517 if (skb->pkt_type == PACKET_LOOPBACK ||
2518 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2522 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2526 return br_handle_frame_hook(port, skb);
2529 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2532 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2533 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2534 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2536 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2537 struct packet_type **pt_prev,
2539 struct net_device *orig_dev)
2541 if (skb->dev->macvlan_port == NULL)
2545 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2548 return macvlan_handle_frame_hook(skb);
2551 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2554 #ifdef CONFIG_NET_CLS_ACT
2555 /* TODO: Maybe we should just force sch_ingress to be compiled in
2556 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2557 * a compare and 2 stores extra right now if we dont have it on
2558 * but have CONFIG_NET_CLS_ACT
2559 * NOTE: This doesnt stop any functionality; if you dont have
2560 * the ingress scheduler, you just cant add policies on ingress.
2563 static int ing_filter(struct sk_buff *skb)
2565 struct net_device *dev = skb->dev;
2566 u32 ttl = G_TC_RTTL(skb->tc_verd);
2567 struct netdev_queue *rxq;
2568 int result = TC_ACT_OK;
2571 if (MAX_RED_LOOP < ttl++) {
2573 "Redir loop detected Dropping packet (%d->%d)\n",
2574 skb->skb_iif, dev->ifindex);
2578 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2579 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2581 rxq = &dev->rx_queue;
2584 if (q != &noop_qdisc) {
2585 spin_lock(qdisc_lock(q));
2586 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2587 result = qdisc_enqueue_root(skb, q);
2588 spin_unlock(qdisc_lock(q));
2594 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2595 struct packet_type **pt_prev,
2596 int *ret, struct net_device *orig_dev)
2598 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2602 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2605 /* Huh? Why does turning on AF_PACKET affect this? */
2606 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2609 switch (ing_filter(skb)) {
2623 * netif_nit_deliver - deliver received packets to network taps
2626 * This function is used to deliver incoming packets to network
2627 * taps. It should be used when the normal netif_receive_skb path
2628 * is bypassed, for example because of VLAN acceleration.
2630 void netif_nit_deliver(struct sk_buff *skb)
2632 struct packet_type *ptype;
2634 if (list_empty(&ptype_all))
2637 skb_reset_network_header(skb);
2638 skb_reset_transport_header(skb);
2639 skb->mac_len = skb->network_header - skb->mac_header;
2642 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2643 if (!ptype->dev || ptype->dev == skb->dev)
2644 deliver_skb(skb, ptype, skb->dev);
2649 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2650 struct net_device *master)
2652 if (skb->pkt_type == PACKET_HOST) {
2653 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2655 memcpy(dest, master->dev_addr, ETH_ALEN);
2659 /* On bonding slaves other than the currently active slave, suppress
2660 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2661 * ARP on active-backup slaves with arp_validate enabled.
2663 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2665 struct net_device *dev = skb->dev;
2667 if (master->priv_flags & IFF_MASTER_ARPMON)
2668 dev->last_rx = jiffies;
2670 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2671 /* Do address unmangle. The local destination address
2672 * will be always the one master has. Provides the right
2673 * functionality in a bridge.
2675 skb_bond_set_mac_by_master(skb, master);
2678 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2679 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2680 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2683 if (master->priv_flags & IFF_MASTER_ALB) {
2684 if (skb->pkt_type != PACKET_BROADCAST &&
2685 skb->pkt_type != PACKET_MULTICAST)
2688 if (master->priv_flags & IFF_MASTER_8023AD &&
2689 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2696 EXPORT_SYMBOL(__skb_bond_should_drop);
2698 static int __netif_receive_skb(struct sk_buff *skb)
2700 struct packet_type *ptype, *pt_prev;
2701 struct net_device *orig_dev;
2702 struct net_device *master;
2703 struct net_device *null_or_orig;
2704 struct net_device *null_or_bond;
2705 int ret = NET_RX_DROP;
2708 if (!skb->tstamp.tv64)
2711 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2712 return NET_RX_SUCCESS;
2714 /* if we've gotten here through NAPI, check netpoll */
2715 if (netpoll_receive_skb(skb))
2719 skb->skb_iif = skb->dev->ifindex;
2721 null_or_orig = NULL;
2722 orig_dev = skb->dev;
2723 master = ACCESS_ONCE(orig_dev->master);
2725 if (skb_bond_should_drop(skb, master))
2726 null_or_orig = orig_dev; /* deliver only exact match */
2731 __get_cpu_var(netdev_rx_stat).total++;
2733 skb_reset_network_header(skb);
2734 skb_reset_transport_header(skb);
2735 skb->mac_len = skb->network_header - skb->mac_header;
2741 #ifdef CONFIG_NET_CLS_ACT
2742 if (skb->tc_verd & TC_NCLS) {
2743 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2748 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2749 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2750 ptype->dev == orig_dev) {
2752 ret = deliver_skb(skb, pt_prev, orig_dev);
2757 #ifdef CONFIG_NET_CLS_ACT
2758 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2764 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2767 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2772 * Make sure frames received on VLAN interfaces stacked on
2773 * bonding interfaces still make their way to any base bonding
2774 * device that may have registered for a specific ptype. The
2775 * handler may have to adjust skb->dev and orig_dev.
2777 null_or_bond = NULL;
2778 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2779 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2780 null_or_bond = vlan_dev_real_dev(skb->dev);
2783 type = skb->protocol;
2784 list_for_each_entry_rcu(ptype,
2785 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2786 if (ptype->type == type && (ptype->dev == null_or_orig ||
2787 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2788 ptype->dev == null_or_bond)) {
2790 ret = deliver_skb(skb, pt_prev, orig_dev);
2796 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2799 /* Jamal, now you will not able to escape explaining
2800 * me how you were going to use this. :-)
2811 * netif_receive_skb - process receive buffer from network
2812 * @skb: buffer to process
2814 * netif_receive_skb() is the main receive data processing function.
2815 * It always succeeds. The buffer may be dropped during processing
2816 * for congestion control or by the protocol layers.
2818 * This function may only be called from softirq context and interrupts
2819 * should be enabled.
2821 * Return values (usually ignored):
2822 * NET_RX_SUCCESS: no congestion
2823 * NET_RX_DROP: packet was dropped
2825 int netif_receive_skb(struct sk_buff *skb)
2830 cpu = get_rps_cpu(skb->dev, skb);
2833 return __netif_receive_skb(skb);
2835 return enqueue_to_backlog(skb, cpu);
2837 return __netif_receive_skb(skb);
2840 EXPORT_SYMBOL(netif_receive_skb);
2842 /* Network device is going away, flush any packets still pending */
2843 static void flush_backlog(void *arg)
2845 struct net_device *dev = arg;
2846 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2847 struct sk_buff *skb, *tmp;
2850 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2851 if (skb->dev == dev) {
2852 __skb_unlink(skb, &queue->input_pkt_queue);
2858 static int napi_gro_complete(struct sk_buff *skb)
2860 struct packet_type *ptype;
2861 __be16 type = skb->protocol;
2862 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2865 if (NAPI_GRO_CB(skb)->count == 1) {
2866 skb_shinfo(skb)->gso_size = 0;
2871 list_for_each_entry_rcu(ptype, head, list) {
2872 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2875 err = ptype->gro_complete(skb);
2881 WARN_ON(&ptype->list == head);
2883 return NET_RX_SUCCESS;
2887 return netif_receive_skb(skb);
2890 static void napi_gro_flush(struct napi_struct *napi)
2892 struct sk_buff *skb, *next;
2894 for (skb = napi->gro_list; skb; skb = next) {
2897 napi_gro_complete(skb);
2900 napi->gro_count = 0;
2901 napi->gro_list = NULL;
2904 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2906 struct sk_buff **pp = NULL;
2907 struct packet_type *ptype;
2908 __be16 type = skb->protocol;
2909 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2912 enum gro_result ret;
2914 if (!(skb->dev->features & NETIF_F_GRO))
2917 if (skb_is_gso(skb) || skb_has_frags(skb))
2921 list_for_each_entry_rcu(ptype, head, list) {
2922 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2925 skb_set_network_header(skb, skb_gro_offset(skb));
2926 mac_len = skb->network_header - skb->mac_header;
2927 skb->mac_len = mac_len;
2928 NAPI_GRO_CB(skb)->same_flow = 0;
2929 NAPI_GRO_CB(skb)->flush = 0;
2930 NAPI_GRO_CB(skb)->free = 0;
2932 pp = ptype->gro_receive(&napi->gro_list, skb);
2937 if (&ptype->list == head)
2940 same_flow = NAPI_GRO_CB(skb)->same_flow;
2941 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2944 struct sk_buff *nskb = *pp;
2948 napi_gro_complete(nskb);
2955 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2959 NAPI_GRO_CB(skb)->count = 1;
2960 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2961 skb->next = napi->gro_list;
2962 napi->gro_list = skb;
2966 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2967 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2969 BUG_ON(skb->end - skb->tail < grow);
2971 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2974 skb->data_len -= grow;
2976 skb_shinfo(skb)->frags[0].page_offset += grow;
2977 skb_shinfo(skb)->frags[0].size -= grow;
2979 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2980 put_page(skb_shinfo(skb)->frags[0].page);
2981 memmove(skb_shinfo(skb)->frags,
2982 skb_shinfo(skb)->frags + 1,
2983 --skb_shinfo(skb)->nr_frags);
2994 EXPORT_SYMBOL(dev_gro_receive);
2997 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3001 if (netpoll_rx_on(skb))
3004 for (p = napi->gro_list; p; p = p->next) {
3005 NAPI_GRO_CB(p)->same_flow =
3006 (p->dev == skb->dev) &&
3007 !compare_ether_header(skb_mac_header(p),
3008 skb_gro_mac_header(skb));
3009 NAPI_GRO_CB(p)->flush = 0;
3012 return dev_gro_receive(napi, skb);
3015 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3019 if (netif_receive_skb(skb))
3024 case GRO_MERGED_FREE:
3035 EXPORT_SYMBOL(napi_skb_finish);
3037 void skb_gro_reset_offset(struct sk_buff *skb)
3039 NAPI_GRO_CB(skb)->data_offset = 0;
3040 NAPI_GRO_CB(skb)->frag0 = NULL;
3041 NAPI_GRO_CB(skb)->frag0_len = 0;
3043 if (skb->mac_header == skb->tail &&
3044 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3045 NAPI_GRO_CB(skb)->frag0 =
3046 page_address(skb_shinfo(skb)->frags[0].page) +
3047 skb_shinfo(skb)->frags[0].page_offset;
3048 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3051 EXPORT_SYMBOL(skb_gro_reset_offset);
3053 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3055 skb_gro_reset_offset(skb);
3057 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3059 EXPORT_SYMBOL(napi_gro_receive);
3061 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3063 __skb_pull(skb, skb_headlen(skb));
3064 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3068 EXPORT_SYMBOL(napi_reuse_skb);
3070 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3072 struct sk_buff *skb = napi->skb;
3075 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3081 EXPORT_SYMBOL(napi_get_frags);
3083 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3089 skb->protocol = eth_type_trans(skb, skb->dev);
3091 if (ret == GRO_HELD)
3092 skb_gro_pull(skb, -ETH_HLEN);
3093 else if (netif_receive_skb(skb))
3098 case GRO_MERGED_FREE:
3099 napi_reuse_skb(napi, skb);
3108 EXPORT_SYMBOL(napi_frags_finish);
3110 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3112 struct sk_buff *skb = napi->skb;
3119 skb_reset_mac_header(skb);
3120 skb_gro_reset_offset(skb);
3122 off = skb_gro_offset(skb);
3123 hlen = off + sizeof(*eth);
3124 eth = skb_gro_header_fast(skb, off);
3125 if (skb_gro_header_hard(skb, hlen)) {
3126 eth = skb_gro_header_slow(skb, hlen, off);
3127 if (unlikely(!eth)) {
3128 napi_reuse_skb(napi, skb);
3134 skb_gro_pull(skb, sizeof(*eth));
3137 * This works because the only protocols we care about don't require
3138 * special handling. We'll fix it up properly at the end.
3140 skb->protocol = eth->h_proto;
3145 EXPORT_SYMBOL(napi_frags_skb);
3147 gro_result_t napi_gro_frags(struct napi_struct *napi)
3149 struct sk_buff *skb = napi_frags_skb(napi);
3154 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3156 EXPORT_SYMBOL(napi_gro_frags);
3158 static int process_backlog(struct napi_struct *napi, int quota)
3161 struct softnet_data *queue = &__get_cpu_var(softnet_data);
3162 unsigned long start_time = jiffies;
3164 napi->weight = weight_p;
3166 struct sk_buff *skb;
3168 local_irq_disable();
3170 skb = __skb_dequeue(&queue->input_pkt_queue);
3172 __napi_complete(napi);
3180 __netif_receive_skb(skb);
3181 } while (++work < quota && jiffies == start_time);
3187 * __napi_schedule - schedule for receive
3188 * @n: entry to schedule
3190 * The entry's receive function will be scheduled to run
3192 void __napi_schedule(struct napi_struct *n)
3194 unsigned long flags;
3196 local_irq_save(flags);
3197 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
3198 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3199 local_irq_restore(flags);
3201 EXPORT_SYMBOL(__napi_schedule);
3203 void __napi_complete(struct napi_struct *n)
3205 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3206 BUG_ON(n->gro_list);
3208 list_del(&n->poll_list);
3209 smp_mb__before_clear_bit();
3210 clear_bit(NAPI_STATE_SCHED, &n->state);
3212 EXPORT_SYMBOL(__napi_complete);
3214 void napi_complete(struct napi_struct *n)
3216 unsigned long flags;
3219 * don't let napi dequeue from the cpu poll list
3220 * just in case its running on a different cpu
3222 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3226 local_irq_save(flags);
3228 local_irq_restore(flags);
3230 EXPORT_SYMBOL(napi_complete);
3232 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3233 int (*poll)(struct napi_struct *, int), int weight)
3235 INIT_LIST_HEAD(&napi->poll_list);
3236 napi->gro_count = 0;
3237 napi->gro_list = NULL;
3240 napi->weight = weight;
3241 list_add(&napi->dev_list, &dev->napi_list);
3243 #ifdef CONFIG_NETPOLL
3244 spin_lock_init(&napi->poll_lock);
3245 napi->poll_owner = -1;
3247 set_bit(NAPI_STATE_SCHED, &napi->state);
3249 EXPORT_SYMBOL(netif_napi_add);
3251 void netif_napi_del(struct napi_struct *napi)
3253 struct sk_buff *skb, *next;
3255 list_del_init(&napi->dev_list);
3256 napi_free_frags(napi);
3258 for (skb = napi->gro_list; skb; skb = next) {
3264 napi->gro_list = NULL;
3265 napi->gro_count = 0;
3267 EXPORT_SYMBOL(netif_napi_del);
3271 * net_rps_action sends any pending IPI's for rps. This is only called from
3272 * softirq and interrupts must be enabled.
3274 static void net_rps_action(cpumask_t *mask)
3278 /* Send pending IPI's to kick RPS processing on remote cpus. */
3279 for_each_cpu_mask_nr(cpu, *mask) {
3280 struct softnet_data *queue = &per_cpu(softnet_data, cpu);
3281 if (cpu_online(cpu))
3282 __smp_call_function_single(cpu, &queue->csd, 0);
3288 static void net_rx_action(struct softirq_action *h)
3290 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3291 unsigned long time_limit = jiffies + 2;
3292 int budget = netdev_budget;
3296 struct rps_remote_softirq_cpus *rcpus;
3299 local_irq_disable();
3301 while (!list_empty(list)) {
3302 struct napi_struct *n;
3305 /* If softirq window is exhuasted then punt.
3306 * Allow this to run for 2 jiffies since which will allow
3307 * an average latency of 1.5/HZ.
3309 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3314 /* Even though interrupts have been re-enabled, this
3315 * access is safe because interrupts can only add new
3316 * entries to the tail of this list, and only ->poll()
3317 * calls can remove this head entry from the list.
3319 n = list_first_entry(list, struct napi_struct, poll_list);
3321 have = netpoll_poll_lock(n);
3325 /* This NAPI_STATE_SCHED test is for avoiding a race
3326 * with netpoll's poll_napi(). Only the entity which
3327 * obtains the lock and sees NAPI_STATE_SCHED set will
3328 * actually make the ->poll() call. Therefore we avoid
3329 * accidently calling ->poll() when NAPI is not scheduled.
3332 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3333 work = n->poll(n, weight);
3337 WARN_ON_ONCE(work > weight);
3341 local_irq_disable();
3343 /* Drivers must not modify the NAPI state if they
3344 * consume the entire weight. In such cases this code
3345 * still "owns" the NAPI instance and therefore can
3346 * move the instance around on the list at-will.
3348 if (unlikely(work == weight)) {
3349 if (unlikely(napi_disable_pending(n))) {
3352 local_irq_disable();
3354 list_move_tail(&n->poll_list, list);
3357 netpoll_poll_unlock(have);
3361 rcpus = &__get_cpu_var(rps_remote_softirq_cpus);
3362 select = rcpus->select;
3367 net_rps_action(&rcpus->mask[select]);
3372 #ifdef CONFIG_NET_DMA
3374 * There may not be any more sk_buffs coming right now, so push
3375 * any pending DMA copies to hardware
3377 dma_issue_pending_all();
3383 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3384 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3388 static gifconf_func_t *gifconf_list[NPROTO];
3391 * register_gifconf - register a SIOCGIF handler
3392 * @family: Address family
3393 * @gifconf: Function handler
3395 * Register protocol dependent address dumping routines. The handler
3396 * that is passed must not be freed or reused until it has been replaced
3397 * by another handler.
3399 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3401 if (family >= NPROTO)
3403 gifconf_list[family] = gifconf;
3406 EXPORT_SYMBOL(register_gifconf);
3410 * Map an interface index to its name (SIOCGIFNAME)
3414 * We need this ioctl for efficient implementation of the
3415 * if_indextoname() function required by the IPv6 API. Without
3416 * it, we would have to search all the interfaces to find a
3420 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3422 struct net_device *dev;
3426 * Fetch the caller's info block.
3429 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3433 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3439 strcpy(ifr.ifr_name, dev->name);
3442 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3448 * Perform a SIOCGIFCONF call. This structure will change
3449 * size eventually, and there is nothing I can do about it.
3450 * Thus we will need a 'compatibility mode'.
3453 static int dev_ifconf(struct net *net, char __user *arg)
3456 struct net_device *dev;
3463 * Fetch the caller's info block.
3466 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3473 * Loop over the interfaces, and write an info block for each.
3477 for_each_netdev(net, dev) {
3478 for (i = 0; i < NPROTO; i++) {
3479 if (gifconf_list[i]) {
3482 done = gifconf_list[i](dev, NULL, 0);
3484 done = gifconf_list[i](dev, pos + total,
3494 * All done. Write the updated control block back to the caller.
3496 ifc.ifc_len = total;
3499 * Both BSD and Solaris return 0 here, so we do too.
3501 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3504 #ifdef CONFIG_PROC_FS
3506 * This is invoked by the /proc filesystem handler to display a device
3509 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3512 struct net *net = seq_file_net(seq);
3514 struct net_device *dev;
3518 return SEQ_START_TOKEN;
3521 for_each_netdev_rcu(net, dev)
3528 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3530 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3531 first_net_device(seq_file_net(seq)) :
3532 next_net_device((struct net_device *)v);
3535 return rcu_dereference(dev);
3538 void dev_seq_stop(struct seq_file *seq, void *v)
3544 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3546 const struct net_device_stats *stats = dev_get_stats(dev);
3548 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3549 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3550 dev->name, stats->rx_bytes, stats->rx_packets,
3552 stats->rx_dropped + stats->rx_missed_errors,
3553 stats->rx_fifo_errors,
3554 stats->rx_length_errors + stats->rx_over_errors +
3555 stats->rx_crc_errors + stats->rx_frame_errors,
3556 stats->rx_compressed, stats->multicast,
3557 stats->tx_bytes, stats->tx_packets,
3558 stats->tx_errors, stats->tx_dropped,
3559 stats->tx_fifo_errors, stats->collisions,
3560 stats->tx_carrier_errors +
3561 stats->tx_aborted_errors +
3562 stats->tx_window_errors +
3563 stats->tx_heartbeat_errors,
3564 stats->tx_compressed);
3568 * Called from the PROCfs module. This now uses the new arbitrary sized
3569 * /proc/net interface to create /proc/net/dev
3571 static int dev_seq_show(struct seq_file *seq, void *v)
3573 if (v == SEQ_START_TOKEN)
3574 seq_puts(seq, "Inter-| Receive "
3576 " face |bytes packets errs drop fifo frame "
3577 "compressed multicast|bytes packets errs "
3578 "drop fifo colls carrier compressed\n");
3580 dev_seq_printf_stats(seq, v);
3584 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3586 struct netif_rx_stats *rc = NULL;
3588 while (*pos < nr_cpu_ids)
3589 if (cpu_online(*pos)) {
3590 rc = &per_cpu(netdev_rx_stat, *pos);
3597 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3599 return softnet_get_online(pos);
3602 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3605 return softnet_get_online(pos);
3608 static void softnet_seq_stop(struct seq_file *seq, void *v)
3612 static int softnet_seq_show(struct seq_file *seq, void *v)
3614 struct netif_rx_stats *s = v;
3616 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3617 s->total, s->dropped, s->time_squeeze, 0,
3618 0, 0, 0, 0, /* was fastroute */
3619 s->cpu_collision, s->received_rps);
3623 static const struct seq_operations dev_seq_ops = {
3624 .start = dev_seq_start,
3625 .next = dev_seq_next,
3626 .stop = dev_seq_stop,
3627 .show = dev_seq_show,
3630 static int dev_seq_open(struct inode *inode, struct file *file)
3632 return seq_open_net(inode, file, &dev_seq_ops,
3633 sizeof(struct seq_net_private));
3636 static const struct file_operations dev_seq_fops = {
3637 .owner = THIS_MODULE,
3638 .open = dev_seq_open,
3640 .llseek = seq_lseek,
3641 .release = seq_release_net,
3644 static const struct seq_operations softnet_seq_ops = {
3645 .start = softnet_seq_start,
3646 .next = softnet_seq_next,
3647 .stop = softnet_seq_stop,
3648 .show = softnet_seq_show,
3651 static int softnet_seq_open(struct inode *inode, struct file *file)
3653 return seq_open(file, &softnet_seq_ops);
3656 static const struct file_operations softnet_seq_fops = {
3657 .owner = THIS_MODULE,
3658 .open = softnet_seq_open,
3660 .llseek = seq_lseek,
3661 .release = seq_release,
3664 static void *ptype_get_idx(loff_t pos)
3666 struct packet_type *pt = NULL;
3670 list_for_each_entry_rcu(pt, &ptype_all, list) {
3676 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3677 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3686 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3690 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3693 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3695 struct packet_type *pt;
3696 struct list_head *nxt;
3700 if (v == SEQ_START_TOKEN)
3701 return ptype_get_idx(0);
3704 nxt = pt->list.next;
3705 if (pt->type == htons(ETH_P_ALL)) {
3706 if (nxt != &ptype_all)
3709 nxt = ptype_base[0].next;
3711 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3713 while (nxt == &ptype_base[hash]) {
3714 if (++hash >= PTYPE_HASH_SIZE)
3716 nxt = ptype_base[hash].next;
3719 return list_entry(nxt, struct packet_type, list);
3722 static void ptype_seq_stop(struct seq_file *seq, void *v)
3728 static int ptype_seq_show(struct seq_file *seq, void *v)
3730 struct packet_type *pt = v;
3732 if (v == SEQ_START_TOKEN)
3733 seq_puts(seq, "Type Device Function\n");
3734 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3735 if (pt->type == htons(ETH_P_ALL))
3736 seq_puts(seq, "ALL ");
3738 seq_printf(seq, "%04x", ntohs(pt->type));
3740 seq_printf(seq, " %-8s %pF\n",
3741 pt->dev ? pt->dev->name : "", pt->func);
3747 static const struct seq_operations ptype_seq_ops = {
3748 .start = ptype_seq_start,
3749 .next = ptype_seq_next,
3750 .stop = ptype_seq_stop,
3751 .show = ptype_seq_show,
3754 static int ptype_seq_open(struct inode *inode, struct file *file)
3756 return seq_open_net(inode, file, &ptype_seq_ops,
3757 sizeof(struct seq_net_private));
3760 static const struct file_operations ptype_seq_fops = {
3761 .owner = THIS_MODULE,
3762 .open = ptype_seq_open,
3764 .llseek = seq_lseek,
3765 .release = seq_release_net,
3769 static int __net_init dev_proc_net_init(struct net *net)
3773 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3775 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3777 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3780 if (wext_proc_init(net))
3786 proc_net_remove(net, "ptype");
3788 proc_net_remove(net, "softnet_stat");
3790 proc_net_remove(net, "dev");
3794 static void __net_exit dev_proc_net_exit(struct net *net)
3796 wext_proc_exit(net);
3798 proc_net_remove(net, "ptype");
3799 proc_net_remove(net, "softnet_stat");
3800 proc_net_remove(net, "dev");
3803 static struct pernet_operations __net_initdata dev_proc_ops = {
3804 .init = dev_proc_net_init,
3805 .exit = dev_proc_net_exit,
3808 static int __init dev_proc_init(void)
3810 return register_pernet_subsys(&dev_proc_ops);
3813 #define dev_proc_init() 0
3814 #endif /* CONFIG_PROC_FS */
3818 * netdev_set_master - set up master/slave pair
3819 * @slave: slave device
3820 * @master: new master device
3822 * Changes the master device of the slave. Pass %NULL to break the
3823 * bonding. The caller must hold the RTNL semaphore. On a failure
3824 * a negative errno code is returned. On success the reference counts
3825 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3826 * function returns zero.
3828 int netdev_set_master(struct net_device *slave, struct net_device *master)
3830 struct net_device *old = slave->master;
3840 slave->master = master;
3847 slave->flags |= IFF_SLAVE;
3849 slave->flags &= ~IFF_SLAVE;
3851 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3854 EXPORT_SYMBOL(netdev_set_master);
3856 static void dev_change_rx_flags(struct net_device *dev, int flags)
3858 const struct net_device_ops *ops = dev->netdev_ops;
3860 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3861 ops->ndo_change_rx_flags(dev, flags);
3864 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3866 unsigned short old_flags = dev->flags;
3872 dev->flags |= IFF_PROMISC;
3873 dev->promiscuity += inc;
3874 if (dev->promiscuity == 0) {
3877 * If inc causes overflow, untouch promisc and return error.
3880 dev->flags &= ~IFF_PROMISC;
3882 dev->promiscuity -= inc;
3883 printk(KERN_WARNING "%s: promiscuity touches roof, "
3884 "set promiscuity failed, promiscuity feature "
3885 "of device might be broken.\n", dev->name);
3889 if (dev->flags != old_flags) {
3890 printk(KERN_INFO "device %s %s promiscuous mode\n",
3891 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3893 if (audit_enabled) {
3894 current_uid_gid(&uid, &gid);
3895 audit_log(current->audit_context, GFP_ATOMIC,
3896 AUDIT_ANOM_PROMISCUOUS,
3897 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3898 dev->name, (dev->flags & IFF_PROMISC),
3899 (old_flags & IFF_PROMISC),
3900 audit_get_loginuid(current),
3902 audit_get_sessionid(current));
3905 dev_change_rx_flags(dev, IFF_PROMISC);
3911 * dev_set_promiscuity - update promiscuity count on a device
3915 * Add or remove promiscuity from a device. While the count in the device
3916 * remains above zero the interface remains promiscuous. Once it hits zero
3917 * the device reverts back to normal filtering operation. A negative inc
3918 * value is used to drop promiscuity on the device.
3919 * Return 0 if successful or a negative errno code on error.
3921 int dev_set_promiscuity(struct net_device *dev, int inc)
3923 unsigned short old_flags = dev->flags;
3926 err = __dev_set_promiscuity(dev, inc);
3929 if (dev->flags != old_flags)
3930 dev_set_rx_mode(dev);
3933 EXPORT_SYMBOL(dev_set_promiscuity);
3936 * dev_set_allmulti - update allmulti count on a device
3940 * Add or remove reception of all multicast frames to a device. While the
3941 * count in the device remains above zero the interface remains listening
3942 * to all interfaces. Once it hits zero the device reverts back to normal
3943 * filtering operation. A negative @inc value is used to drop the counter
3944 * when releasing a resource needing all multicasts.
3945 * Return 0 if successful or a negative errno code on error.
3948 int dev_set_allmulti(struct net_device *dev, int inc)
3950 unsigned short old_flags = dev->flags;
3954 dev->flags |= IFF_ALLMULTI;
3955 dev->allmulti += inc;
3956 if (dev->allmulti == 0) {
3959 * If inc causes overflow, untouch allmulti and return error.
3962 dev->flags &= ~IFF_ALLMULTI;
3964 dev->allmulti -= inc;
3965 printk(KERN_WARNING "%s: allmulti touches roof, "
3966 "set allmulti failed, allmulti feature of "
3967 "device might be broken.\n", dev->name);
3971 if (dev->flags ^ old_flags) {
3972 dev_change_rx_flags(dev, IFF_ALLMULTI);
3973 dev_set_rx_mode(dev);
3977 EXPORT_SYMBOL(dev_set_allmulti);
3980 * Upload unicast and multicast address lists to device and
3981 * configure RX filtering. When the device doesn't support unicast
3982 * filtering it is put in promiscuous mode while unicast addresses
3985 void __dev_set_rx_mode(struct net_device *dev)
3987 const struct net_device_ops *ops = dev->netdev_ops;
3989 /* dev_open will call this function so the list will stay sane. */
3990 if (!(dev->flags&IFF_UP))
3993 if (!netif_device_present(dev))
3996 if (ops->ndo_set_rx_mode)
3997 ops->ndo_set_rx_mode(dev);
3999 /* Unicast addresses changes may only happen under the rtnl,
4000 * therefore calling __dev_set_promiscuity here is safe.
4002 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4003 __dev_set_promiscuity(dev, 1);
4004 dev->uc_promisc = 1;
4005 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4006 __dev_set_promiscuity(dev, -1);
4007 dev->uc_promisc = 0;
4010 if (ops->ndo_set_multicast_list)
4011 ops->ndo_set_multicast_list(dev);
4015 void dev_set_rx_mode(struct net_device *dev)
4017 netif_addr_lock_bh(dev);
4018 __dev_set_rx_mode(dev);
4019 netif_addr_unlock_bh(dev);
4023 * dev_get_flags - get flags reported to userspace
4026 * Get the combination of flag bits exported through APIs to userspace.
4028 unsigned dev_get_flags(const struct net_device *dev)
4032 flags = (dev->flags & ~(IFF_PROMISC |
4037 (dev->gflags & (IFF_PROMISC |
4040 if (netif_running(dev)) {
4041 if (netif_oper_up(dev))
4042 flags |= IFF_RUNNING;
4043 if (netif_carrier_ok(dev))
4044 flags |= IFF_LOWER_UP;
4045 if (netif_dormant(dev))
4046 flags |= IFF_DORMANT;
4051 EXPORT_SYMBOL(dev_get_flags);
4053 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4055 int old_flags = dev->flags;
4061 * Set the flags on our device.
4064 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4065 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4067 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4071 * Load in the correct multicast list now the flags have changed.
4074 if ((old_flags ^ flags) & IFF_MULTICAST)
4075 dev_change_rx_flags(dev, IFF_MULTICAST);
4077 dev_set_rx_mode(dev);
4080 * Have we downed the interface. We handle IFF_UP ourselves
4081 * according to user attempts to set it, rather than blindly
4086 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4087 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4090 dev_set_rx_mode(dev);
4093 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4094 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4096 dev->gflags ^= IFF_PROMISC;
4097 dev_set_promiscuity(dev, inc);
4100 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4101 is important. Some (broken) drivers set IFF_PROMISC, when
4102 IFF_ALLMULTI is requested not asking us and not reporting.
4104 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4105 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4107 dev->gflags ^= IFF_ALLMULTI;
4108 dev_set_allmulti(dev, inc);
4114 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4116 unsigned int changes = dev->flags ^ old_flags;
4118 if (changes & IFF_UP) {
4119 if (dev->flags & IFF_UP)
4120 call_netdevice_notifiers(NETDEV_UP, dev);
4122 call_netdevice_notifiers(NETDEV_DOWN, dev);
4125 if (dev->flags & IFF_UP &&
4126 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4127 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4131 * dev_change_flags - change device settings
4133 * @flags: device state flags
4135 * Change settings on device based state flags. The flags are
4136 * in the userspace exported format.
4138 int dev_change_flags(struct net_device *dev, unsigned flags)
4141 int old_flags = dev->flags;
4143 ret = __dev_change_flags(dev, flags);
4147 changes = old_flags ^ dev->flags;
4149 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4151 __dev_notify_flags(dev, old_flags);
4154 EXPORT_SYMBOL(dev_change_flags);
4157 * dev_set_mtu - Change maximum transfer unit
4159 * @new_mtu: new transfer unit
4161 * Change the maximum transfer size of the network device.
4163 int dev_set_mtu(struct net_device *dev, int new_mtu)
4165 const struct net_device_ops *ops = dev->netdev_ops;
4168 if (new_mtu == dev->mtu)
4171 /* MTU must be positive. */
4175 if (!netif_device_present(dev))
4179 if (ops->ndo_change_mtu)
4180 err = ops->ndo_change_mtu(dev, new_mtu);
4184 if (!err && dev->flags & IFF_UP)
4185 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4188 EXPORT_SYMBOL(dev_set_mtu);
4191 * dev_set_mac_address - Change Media Access Control Address
4195 * Change the hardware (MAC) address of the device
4197 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4199 const struct net_device_ops *ops = dev->netdev_ops;
4202 if (!ops->ndo_set_mac_address)
4204 if (sa->sa_family != dev->type)
4206 if (!netif_device_present(dev))
4208 err = ops->ndo_set_mac_address(dev, sa);
4210 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4213 EXPORT_SYMBOL(dev_set_mac_address);
4216 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4218 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4221 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4227 case SIOCGIFFLAGS: /* Get interface flags */
4228 ifr->ifr_flags = (short) dev_get_flags(dev);
4231 case SIOCGIFMETRIC: /* Get the metric on the interface
4232 (currently unused) */
4233 ifr->ifr_metric = 0;
4236 case SIOCGIFMTU: /* Get the MTU of a device */
4237 ifr->ifr_mtu = dev->mtu;
4242 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4244 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4245 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4246 ifr->ifr_hwaddr.sa_family = dev->type;
4254 ifr->ifr_map.mem_start = dev->mem_start;
4255 ifr->ifr_map.mem_end = dev->mem_end;
4256 ifr->ifr_map.base_addr = dev->base_addr;
4257 ifr->ifr_map.irq = dev->irq;
4258 ifr->ifr_map.dma = dev->dma;
4259 ifr->ifr_map.port = dev->if_port;
4263 ifr->ifr_ifindex = dev->ifindex;
4267 ifr->ifr_qlen = dev->tx_queue_len;
4271 /* dev_ioctl() should ensure this case
4283 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4285 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4288 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4289 const struct net_device_ops *ops;
4294 ops = dev->netdev_ops;
4297 case SIOCSIFFLAGS: /* Set interface flags */
4298 return dev_change_flags(dev, ifr->ifr_flags);
4300 case SIOCSIFMETRIC: /* Set the metric on the interface
4301 (currently unused) */
4304 case SIOCSIFMTU: /* Set the MTU of a device */
4305 return dev_set_mtu(dev, ifr->ifr_mtu);
4308 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4310 case SIOCSIFHWBROADCAST:
4311 if (ifr->ifr_hwaddr.sa_family != dev->type)
4313 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4314 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4315 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4319 if (ops->ndo_set_config) {
4320 if (!netif_device_present(dev))
4322 return ops->ndo_set_config(dev, &ifr->ifr_map);
4327 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4328 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4330 if (!netif_device_present(dev))
4332 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4335 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4336 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4338 if (!netif_device_present(dev))
4340 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4343 if (ifr->ifr_qlen < 0)
4345 dev->tx_queue_len = ifr->ifr_qlen;
4349 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4350 return dev_change_name(dev, ifr->ifr_newname);
4353 * Unknown or private ioctl
4356 if ((cmd >= SIOCDEVPRIVATE &&
4357 cmd <= SIOCDEVPRIVATE + 15) ||
4358 cmd == SIOCBONDENSLAVE ||
4359 cmd == SIOCBONDRELEASE ||
4360 cmd == SIOCBONDSETHWADDR ||
4361 cmd == SIOCBONDSLAVEINFOQUERY ||
4362 cmd == SIOCBONDINFOQUERY ||
4363 cmd == SIOCBONDCHANGEACTIVE ||
4364 cmd == SIOCGMIIPHY ||
4365 cmd == SIOCGMIIREG ||
4366 cmd == SIOCSMIIREG ||
4367 cmd == SIOCBRADDIF ||
4368 cmd == SIOCBRDELIF ||
4369 cmd == SIOCSHWTSTAMP ||
4370 cmd == SIOCWANDEV) {
4372 if (ops->ndo_do_ioctl) {
4373 if (netif_device_present(dev))
4374 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4386 * This function handles all "interface"-type I/O control requests. The actual
4387 * 'doing' part of this is dev_ifsioc above.
4391 * dev_ioctl - network device ioctl
4392 * @net: the applicable net namespace
4393 * @cmd: command to issue
4394 * @arg: pointer to a struct ifreq in user space
4396 * Issue ioctl functions to devices. This is normally called by the
4397 * user space syscall interfaces but can sometimes be useful for
4398 * other purposes. The return value is the return from the syscall if
4399 * positive or a negative errno code on error.
4402 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4408 /* One special case: SIOCGIFCONF takes ifconf argument
4409 and requires shared lock, because it sleeps writing
4413 if (cmd == SIOCGIFCONF) {
4415 ret = dev_ifconf(net, (char __user *) arg);
4419 if (cmd == SIOCGIFNAME)
4420 return dev_ifname(net, (struct ifreq __user *)arg);
4422 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4425 ifr.ifr_name[IFNAMSIZ-1] = 0;
4427 colon = strchr(ifr.ifr_name, ':');
4432 * See which interface the caller is talking about.
4437 * These ioctl calls:
4438 * - can be done by all.
4439 * - atomic and do not require locking.
4450 dev_load(net, ifr.ifr_name);
4452 ret = dev_ifsioc_locked(net, &ifr, cmd);
4457 if (copy_to_user(arg, &ifr,
4458 sizeof(struct ifreq)))
4464 dev_load(net, ifr.ifr_name);
4466 ret = dev_ethtool(net, &ifr);
4471 if (copy_to_user(arg, &ifr,
4472 sizeof(struct ifreq)))
4478 * These ioctl calls:
4479 * - require superuser power.
4480 * - require strict serialization.
4486 if (!capable(CAP_NET_ADMIN))
4488 dev_load(net, ifr.ifr_name);
4490 ret = dev_ifsioc(net, &ifr, cmd);
4495 if (copy_to_user(arg, &ifr,
4496 sizeof(struct ifreq)))
4502 * These ioctl calls:
4503 * - require superuser power.
4504 * - require strict serialization.
4505 * - do not return a value
4515 case SIOCSIFHWBROADCAST:
4518 case SIOCBONDENSLAVE:
4519 case SIOCBONDRELEASE:
4520 case SIOCBONDSETHWADDR:
4521 case SIOCBONDCHANGEACTIVE:
4525 if (!capable(CAP_NET_ADMIN))
4528 case SIOCBONDSLAVEINFOQUERY:
4529 case SIOCBONDINFOQUERY:
4530 dev_load(net, ifr.ifr_name);
4532 ret = dev_ifsioc(net, &ifr, cmd);
4537 /* Get the per device memory space. We can add this but
4538 * currently do not support it */
4540 /* Set the per device memory buffer space.
4541 * Not applicable in our case */
4546 * Unknown or private ioctl.
4549 if (cmd == SIOCWANDEV ||
4550 (cmd >= SIOCDEVPRIVATE &&
4551 cmd <= SIOCDEVPRIVATE + 15)) {
4552 dev_load(net, ifr.ifr_name);
4554 ret = dev_ifsioc(net, &ifr, cmd);
4556 if (!ret && copy_to_user(arg, &ifr,
4557 sizeof(struct ifreq)))
4561 /* Take care of Wireless Extensions */
4562 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4563 return wext_handle_ioctl(net, &ifr, cmd, arg);
4570 * dev_new_index - allocate an ifindex
4571 * @net: the applicable net namespace
4573 * Returns a suitable unique value for a new device interface
4574 * number. The caller must hold the rtnl semaphore or the
4575 * dev_base_lock to be sure it remains unique.
4577 static int dev_new_index(struct net *net)
4583 if (!__dev_get_by_index(net, ifindex))
4588 /* Delayed registration/unregisteration */
4589 static LIST_HEAD(net_todo_list);
4591 static void net_set_todo(struct net_device *dev)
4593 list_add_tail(&dev->todo_list, &net_todo_list);
4596 static void rollback_registered_many(struct list_head *head)
4598 struct net_device *dev, *tmp;
4600 BUG_ON(dev_boot_phase);
4603 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4604 /* Some devices call without registering
4605 * for initialization unwind. Remove those
4606 * devices and proceed with the remaining.
4608 if (dev->reg_state == NETREG_UNINITIALIZED) {
4609 pr_debug("unregister_netdevice: device %s/%p never "
4610 "was registered\n", dev->name, dev);
4613 list_del(&dev->unreg_list);
4617 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4619 /* If device is running, close it first. */
4622 /* And unlink it from device chain. */
4623 unlist_netdevice(dev);
4625 dev->reg_state = NETREG_UNREGISTERING;
4630 list_for_each_entry(dev, head, unreg_list) {
4631 /* Shutdown queueing discipline. */
4635 /* Notify protocols, that we are about to destroy
4636 this device. They should clean all the things.
4638 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4640 if (!dev->rtnl_link_ops ||
4641 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4642 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4645 * Flush the unicast and multicast chains
4650 if (dev->netdev_ops->ndo_uninit)
4651 dev->netdev_ops->ndo_uninit(dev);
4653 /* Notifier chain MUST detach us from master device. */
4654 WARN_ON(dev->master);
4656 /* Remove entries from kobject tree */
4657 netdev_unregister_kobject(dev);
4660 /* Process any work delayed until the end of the batch */
4661 dev = list_first_entry(head, struct net_device, unreg_list);
4662 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4666 list_for_each_entry(dev, head, unreg_list)
4670 static void rollback_registered(struct net_device *dev)
4674 list_add(&dev->unreg_list, &single);
4675 rollback_registered_many(&single);
4678 static void __netdev_init_queue_locks_one(struct net_device *dev,
4679 struct netdev_queue *dev_queue,
4682 spin_lock_init(&dev_queue->_xmit_lock);
4683 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4684 dev_queue->xmit_lock_owner = -1;
4687 static void netdev_init_queue_locks(struct net_device *dev)
4689 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4690 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4693 unsigned long netdev_fix_features(unsigned long features, const char *name)
4695 /* Fix illegal SG+CSUM combinations. */
4696 if ((features & NETIF_F_SG) &&
4697 !(features & NETIF_F_ALL_CSUM)) {
4699 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4700 "checksum feature.\n", name);
4701 features &= ~NETIF_F_SG;
4704 /* TSO requires that SG is present as well. */
4705 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4707 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4708 "SG feature.\n", name);
4709 features &= ~NETIF_F_TSO;
4712 if (features & NETIF_F_UFO) {
4713 if (!(features & NETIF_F_GEN_CSUM)) {
4715 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4716 "since no NETIF_F_HW_CSUM feature.\n",
4718 features &= ~NETIF_F_UFO;
4721 if (!(features & NETIF_F_SG)) {
4723 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4724 "since no NETIF_F_SG feature.\n", name);
4725 features &= ~NETIF_F_UFO;
4731 EXPORT_SYMBOL(netdev_fix_features);
4734 * netif_stacked_transfer_operstate - transfer operstate
4735 * @rootdev: the root or lower level device to transfer state from
4736 * @dev: the device to transfer operstate to
4738 * Transfer operational state from root to device. This is normally
4739 * called when a stacking relationship exists between the root
4740 * device and the device(a leaf device).
4742 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4743 struct net_device *dev)
4745 if (rootdev->operstate == IF_OPER_DORMANT)
4746 netif_dormant_on(dev);
4748 netif_dormant_off(dev);
4750 if (netif_carrier_ok(rootdev)) {
4751 if (!netif_carrier_ok(dev))
4752 netif_carrier_on(dev);
4754 if (netif_carrier_ok(dev))
4755 netif_carrier_off(dev);
4758 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4761 * register_netdevice - register a network device
4762 * @dev: device to register
4764 * Take a completed network device structure and add it to the kernel
4765 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4766 * chain. 0 is returned on success. A negative errno code is returned
4767 * on a failure to set up the device, or if the name is a duplicate.
4769 * Callers must hold the rtnl semaphore. You may want
4770 * register_netdev() instead of this.
4773 * The locking appears insufficient to guarantee two parallel registers
4774 * will not get the same name.
4777 int register_netdevice(struct net_device *dev)
4780 struct net *net = dev_net(dev);
4782 BUG_ON(dev_boot_phase);
4787 /* When net_device's are persistent, this will be fatal. */
4788 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4791 spin_lock_init(&dev->addr_list_lock);
4792 netdev_set_addr_lockdep_class(dev);
4793 netdev_init_queue_locks(dev);
4798 if (!dev->num_rx_queues) {
4800 * Allocate a single RX queue if driver never called
4804 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
4810 dev->_rx->first = dev->_rx;
4811 atomic_set(&dev->_rx->count, 1);
4812 dev->num_rx_queues = 1;
4815 /* Init, if this function is available */
4816 if (dev->netdev_ops->ndo_init) {
4817 ret = dev->netdev_ops->ndo_init(dev);
4825 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
4829 dev->ifindex = dev_new_index(net);
4830 if (dev->iflink == -1)
4831 dev->iflink = dev->ifindex;
4833 /* Fix illegal checksum combinations */
4834 if ((dev->features & NETIF_F_HW_CSUM) &&
4835 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4836 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4838 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4841 if ((dev->features & NETIF_F_NO_CSUM) &&
4842 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4843 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4845 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4848 dev->features = netdev_fix_features(dev->features, dev->name);
4850 /* Enable software GSO if SG is supported. */
4851 if (dev->features & NETIF_F_SG)
4852 dev->features |= NETIF_F_GSO;
4854 netdev_initialize_kobject(dev);
4856 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4857 ret = notifier_to_errno(ret);
4861 ret = netdev_register_kobject(dev);
4864 dev->reg_state = NETREG_REGISTERED;
4867 * Default initial state at registry is that the
4868 * device is present.
4871 set_bit(__LINK_STATE_PRESENT, &dev->state);
4873 dev_init_scheduler(dev);
4875 list_netdevice(dev);
4877 /* Notify protocols, that a new device appeared. */
4878 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4879 ret = notifier_to_errno(ret);
4881 rollback_registered(dev);
4882 dev->reg_state = NETREG_UNREGISTERED;
4885 * Prevent userspace races by waiting until the network
4886 * device is fully setup before sending notifications.
4888 if (!dev->rtnl_link_ops ||
4889 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4890 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
4896 if (dev->netdev_ops->ndo_uninit)
4897 dev->netdev_ops->ndo_uninit(dev);
4900 EXPORT_SYMBOL(register_netdevice);
4903 * init_dummy_netdev - init a dummy network device for NAPI
4904 * @dev: device to init
4906 * This takes a network device structure and initialize the minimum
4907 * amount of fields so it can be used to schedule NAPI polls without
4908 * registering a full blown interface. This is to be used by drivers
4909 * that need to tie several hardware interfaces to a single NAPI
4910 * poll scheduler due to HW limitations.
4912 int init_dummy_netdev(struct net_device *dev)
4914 /* Clear everything. Note we don't initialize spinlocks
4915 * are they aren't supposed to be taken by any of the
4916 * NAPI code and this dummy netdev is supposed to be
4917 * only ever used for NAPI polls
4919 memset(dev, 0, sizeof(struct net_device));
4921 /* make sure we BUG if trying to hit standard
4922 * register/unregister code path
4924 dev->reg_state = NETREG_DUMMY;
4926 /* initialize the ref count */
4927 atomic_set(&dev->refcnt, 1);
4929 /* NAPI wants this */
4930 INIT_LIST_HEAD(&dev->napi_list);
4932 /* a dummy interface is started by default */
4933 set_bit(__LINK_STATE_PRESENT, &dev->state);
4934 set_bit(__LINK_STATE_START, &dev->state);
4938 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4942 * register_netdev - register a network device
4943 * @dev: device to register
4945 * Take a completed network device structure and add it to the kernel
4946 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4947 * chain. 0 is returned on success. A negative errno code is returned
4948 * on a failure to set up the device, or if the name is a duplicate.
4950 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4951 * and expands the device name if you passed a format string to
4954 int register_netdev(struct net_device *dev)
4961 * If the name is a format string the caller wants us to do a
4964 if (strchr(dev->name, '%')) {
4965 err = dev_alloc_name(dev, dev->name);
4970 err = register_netdevice(dev);
4975 EXPORT_SYMBOL(register_netdev);
4978 * netdev_wait_allrefs - wait until all references are gone.
4980 * This is called when unregistering network devices.
4982 * Any protocol or device that holds a reference should register
4983 * for netdevice notification, and cleanup and put back the
4984 * reference if they receive an UNREGISTER event.
4985 * We can get stuck here if buggy protocols don't correctly
4988 static void netdev_wait_allrefs(struct net_device *dev)
4990 unsigned long rebroadcast_time, warning_time;
4992 linkwatch_forget_dev(dev);
4994 rebroadcast_time = warning_time = jiffies;
4995 while (atomic_read(&dev->refcnt) != 0) {
4996 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4999 /* Rebroadcast unregister notification */
5000 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5001 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5002 * should have already handle it the first time */
5004 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5006 /* We must not have linkwatch events
5007 * pending on unregister. If this
5008 * happens, we simply run the queue
5009 * unscheduled, resulting in a noop
5012 linkwatch_run_queue();
5017 rebroadcast_time = jiffies;
5022 if (time_after(jiffies, warning_time + 10 * HZ)) {
5023 printk(KERN_EMERG "unregister_netdevice: "
5024 "waiting for %s to become free. Usage "
5026 dev->name, atomic_read(&dev->refcnt));
5027 warning_time = jiffies;
5036 * register_netdevice(x1);
5037 * register_netdevice(x2);
5039 * unregister_netdevice(y1);
5040 * unregister_netdevice(y2);
5046 * We are invoked by rtnl_unlock().
5047 * This allows us to deal with problems:
5048 * 1) We can delete sysfs objects which invoke hotplug
5049 * without deadlocking with linkwatch via keventd.
5050 * 2) Since we run with the RTNL semaphore not held, we can sleep
5051 * safely in order to wait for the netdev refcnt to drop to zero.
5053 * We must not return until all unregister events added during
5054 * the interval the lock was held have been completed.
5056 void netdev_run_todo(void)
5058 struct list_head list;
5060 /* Snapshot list, allow later requests */
5061 list_replace_init(&net_todo_list, &list);
5065 while (!list_empty(&list)) {
5066 struct net_device *dev
5067 = list_first_entry(&list, struct net_device, todo_list);
5068 list_del(&dev->todo_list);
5070 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5071 printk(KERN_ERR "network todo '%s' but state %d\n",
5072 dev->name, dev->reg_state);
5077 dev->reg_state = NETREG_UNREGISTERED;
5079 on_each_cpu(flush_backlog, dev, 1);
5081 netdev_wait_allrefs(dev);
5084 BUG_ON(atomic_read(&dev->refcnt));
5085 WARN_ON(dev->ip_ptr);
5086 WARN_ON(dev->ip6_ptr);
5087 WARN_ON(dev->dn_ptr);
5089 if (dev->destructor)
5090 dev->destructor(dev);
5092 /* Free network device */
5093 kobject_put(&dev->dev.kobj);
5098 * dev_txq_stats_fold - fold tx_queues stats
5099 * @dev: device to get statistics from
5100 * @stats: struct net_device_stats to hold results
5102 void dev_txq_stats_fold(const struct net_device *dev,
5103 struct net_device_stats *stats)
5105 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5107 struct netdev_queue *txq;
5109 for (i = 0; i < dev->num_tx_queues; i++) {
5110 txq = netdev_get_tx_queue(dev, i);
5111 tx_bytes += txq->tx_bytes;
5112 tx_packets += txq->tx_packets;
5113 tx_dropped += txq->tx_dropped;
5115 if (tx_bytes || tx_packets || tx_dropped) {
5116 stats->tx_bytes = tx_bytes;
5117 stats->tx_packets = tx_packets;
5118 stats->tx_dropped = tx_dropped;
5121 EXPORT_SYMBOL(dev_txq_stats_fold);
5124 * dev_get_stats - get network device statistics
5125 * @dev: device to get statistics from
5127 * Get network statistics from device. The device driver may provide
5128 * its own method by setting dev->netdev_ops->get_stats; otherwise
5129 * the internal statistics structure is used.
5131 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5133 const struct net_device_ops *ops = dev->netdev_ops;
5135 if (ops->ndo_get_stats)
5136 return ops->ndo_get_stats(dev);
5138 dev_txq_stats_fold(dev, &dev->stats);
5141 EXPORT_SYMBOL(dev_get_stats);
5143 static void netdev_init_one_queue(struct net_device *dev,
5144 struct netdev_queue *queue,
5150 static void netdev_init_queues(struct net_device *dev)
5152 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5153 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5154 spin_lock_init(&dev->tx_global_lock);
5158 * alloc_netdev_mq - allocate network device
5159 * @sizeof_priv: size of private data to allocate space for
5160 * @name: device name format string
5161 * @setup: callback to initialize device
5162 * @queue_count: the number of subqueues to allocate
5164 * Allocates a struct net_device with private data area for driver use
5165 * and performs basic initialization. Also allocates subquue structs
5166 * for each queue on the device at the end of the netdevice.
5168 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5169 void (*setup)(struct net_device *), unsigned int queue_count)
5171 struct netdev_queue *tx;
5172 struct net_device *dev;
5174 struct net_device *p;
5176 struct netdev_rx_queue *rx;
5180 BUG_ON(strlen(name) >= sizeof(dev->name));
5182 alloc_size = sizeof(struct net_device);
5184 /* ensure 32-byte alignment of private area */
5185 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5186 alloc_size += sizeof_priv;
5188 /* ensure 32-byte alignment of whole construct */
5189 alloc_size += NETDEV_ALIGN - 1;
5191 p = kzalloc(alloc_size, GFP_KERNEL);
5193 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5197 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5199 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5205 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5207 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5212 atomic_set(&rx->count, queue_count);
5215 * Set a pointer to first element in the array which holds the
5218 for (i = 0; i < queue_count; i++)
5222 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5223 dev->padded = (char *)dev - (char *)p;
5225 if (dev_addr_init(dev))
5231 dev_net_set(dev, &init_net);
5234 dev->num_tx_queues = queue_count;
5235 dev->real_num_tx_queues = queue_count;
5239 dev->num_rx_queues = queue_count;
5242 dev->gso_max_size = GSO_MAX_SIZE;
5244 netdev_init_queues(dev);
5246 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5247 dev->ethtool_ntuple_list.count = 0;
5248 INIT_LIST_HEAD(&dev->napi_list);
5249 INIT_LIST_HEAD(&dev->unreg_list);
5250 INIT_LIST_HEAD(&dev->link_watch_list);
5251 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5253 strcpy(dev->name, name);
5266 EXPORT_SYMBOL(alloc_netdev_mq);
5269 * free_netdev - free network device
5272 * This function does the last stage of destroying an allocated device
5273 * interface. The reference to the device object is released.
5274 * If this is the last reference then it will be freed.
5276 void free_netdev(struct net_device *dev)
5278 struct napi_struct *p, *n;
5280 release_net(dev_net(dev));
5284 /* Flush device addresses */
5285 dev_addr_flush(dev);
5287 /* Clear ethtool n-tuple list */
5288 ethtool_ntuple_flush(dev);
5290 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5293 /* Compatibility with error handling in drivers */
5294 if (dev->reg_state == NETREG_UNINITIALIZED) {
5295 kfree((char *)dev - dev->padded);
5299 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5300 dev->reg_state = NETREG_RELEASED;
5302 /* will free via device release */
5303 put_device(&dev->dev);
5305 EXPORT_SYMBOL(free_netdev);
5308 * synchronize_net - Synchronize with packet receive processing
5310 * Wait for packets currently being received to be done.
5311 * Does not block later packets from starting.
5313 void synchronize_net(void)
5318 EXPORT_SYMBOL(synchronize_net);
5321 * unregister_netdevice_queue - remove device from the kernel
5325 * This function shuts down a device interface and removes it
5326 * from the kernel tables.
5327 * If head not NULL, device is queued to be unregistered later.
5329 * Callers must hold the rtnl semaphore. You may want
5330 * unregister_netdev() instead of this.
5333 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5338 list_move_tail(&dev->unreg_list, head);
5340 rollback_registered(dev);
5341 /* Finish processing unregister after unlock */
5345 EXPORT_SYMBOL(unregister_netdevice_queue);
5348 * unregister_netdevice_many - unregister many devices
5349 * @head: list of devices
5351 void unregister_netdevice_many(struct list_head *head)
5353 struct net_device *dev;
5355 if (!list_empty(head)) {
5356 rollback_registered_many(head);
5357 list_for_each_entry(dev, head, unreg_list)
5361 EXPORT_SYMBOL(unregister_netdevice_many);
5364 * unregister_netdev - remove device from the kernel
5367 * This function shuts down a device interface and removes it
5368 * from the kernel tables.
5370 * This is just a wrapper for unregister_netdevice that takes
5371 * the rtnl semaphore. In general you want to use this and not
5372 * unregister_netdevice.
5374 void unregister_netdev(struct net_device *dev)
5377 unregister_netdevice(dev);
5380 EXPORT_SYMBOL(unregister_netdev);
5383 * dev_change_net_namespace - move device to different nethost namespace
5385 * @net: network namespace
5386 * @pat: If not NULL name pattern to try if the current device name
5387 * is already taken in the destination network namespace.
5389 * This function shuts down a device interface and moves it
5390 * to a new network namespace. On success 0 is returned, on
5391 * a failure a netagive errno code is returned.
5393 * Callers must hold the rtnl semaphore.
5396 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5402 /* Don't allow namespace local devices to be moved. */
5404 if (dev->features & NETIF_F_NETNS_LOCAL)
5408 /* Don't allow real devices to be moved when sysfs
5412 if (dev->dev.parent)
5416 /* Ensure the device has been registrered */
5418 if (dev->reg_state != NETREG_REGISTERED)
5421 /* Get out if there is nothing todo */
5423 if (net_eq(dev_net(dev), net))
5426 /* Pick the destination device name, and ensure
5427 * we can use it in the destination network namespace.
5430 if (__dev_get_by_name(net, dev->name)) {
5431 /* We get here if we can't use the current device name */
5434 if (dev_get_valid_name(net, pat, dev->name, 1))
5439 * And now a mini version of register_netdevice unregister_netdevice.
5442 /* If device is running close it first. */
5445 /* And unlink it from device chain */
5447 unlist_netdevice(dev);
5451 /* Shutdown queueing discipline. */
5454 /* Notify protocols, that we are about to destroy
5455 this device. They should clean all the things.
5457 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5458 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5461 * Flush the unicast and multicast chains
5466 netdev_unregister_kobject(dev);
5468 /* Actually switch the network namespace */
5469 dev_net_set(dev, net);
5471 /* If there is an ifindex conflict assign a new one */
5472 if (__dev_get_by_index(net, dev->ifindex)) {
5473 int iflink = (dev->iflink == dev->ifindex);
5474 dev->ifindex = dev_new_index(net);
5476 dev->iflink = dev->ifindex;
5479 /* Fixup kobjects */
5480 err = netdev_register_kobject(dev);
5483 /* Add the device back in the hashes */
5484 list_netdevice(dev);
5486 /* Notify protocols, that a new device appeared. */
5487 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5490 * Prevent userspace races by waiting until the network
5491 * device is fully setup before sending notifications.
5493 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5500 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5502 static int dev_cpu_callback(struct notifier_block *nfb,
5503 unsigned long action,
5506 struct sk_buff **list_skb;
5507 struct Qdisc **list_net;
5508 struct sk_buff *skb;
5509 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5510 struct softnet_data *sd, *oldsd;
5512 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5515 local_irq_disable();
5516 cpu = smp_processor_id();
5517 sd = &per_cpu(softnet_data, cpu);
5518 oldsd = &per_cpu(softnet_data, oldcpu);
5520 /* Find end of our completion_queue. */
5521 list_skb = &sd->completion_queue;
5523 list_skb = &(*list_skb)->next;
5524 /* Append completion queue from offline CPU. */
5525 *list_skb = oldsd->completion_queue;
5526 oldsd->completion_queue = NULL;
5528 /* Find end of our output_queue. */
5529 list_net = &sd->output_queue;
5531 list_net = &(*list_net)->next_sched;
5532 /* Append output queue from offline CPU. */
5533 *list_net = oldsd->output_queue;
5534 oldsd->output_queue = NULL;
5536 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5539 /* Process offline CPU's input_pkt_queue */
5540 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5548 * netdev_increment_features - increment feature set by one
5549 * @all: current feature set
5550 * @one: new feature set
5551 * @mask: mask feature set
5553 * Computes a new feature set after adding a device with feature set
5554 * @one to the master device with current feature set @all. Will not
5555 * enable anything that is off in @mask. Returns the new feature set.
5557 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5560 /* If device needs checksumming, downgrade to it. */
5561 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5562 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5563 else if (mask & NETIF_F_ALL_CSUM) {
5564 /* If one device supports v4/v6 checksumming, set for all. */
5565 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5566 !(all & NETIF_F_GEN_CSUM)) {
5567 all &= ~NETIF_F_ALL_CSUM;
5568 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5571 /* If one device supports hw checksumming, set for all. */
5572 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5573 all &= ~NETIF_F_ALL_CSUM;
5574 all |= NETIF_F_HW_CSUM;
5578 one |= NETIF_F_ALL_CSUM;
5580 one |= all & NETIF_F_ONE_FOR_ALL;
5581 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5582 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5586 EXPORT_SYMBOL(netdev_increment_features);
5588 static struct hlist_head *netdev_create_hash(void)
5591 struct hlist_head *hash;
5593 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5595 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5596 INIT_HLIST_HEAD(&hash[i]);
5601 /* Initialize per network namespace state */
5602 static int __net_init netdev_init(struct net *net)
5604 INIT_LIST_HEAD(&net->dev_base_head);
5606 net->dev_name_head = netdev_create_hash();
5607 if (net->dev_name_head == NULL)
5610 net->dev_index_head = netdev_create_hash();
5611 if (net->dev_index_head == NULL)
5617 kfree(net->dev_name_head);
5623 * netdev_drivername - network driver for the device
5624 * @dev: network device
5625 * @buffer: buffer for resulting name
5626 * @len: size of buffer
5628 * Determine network driver for device.
5630 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5632 const struct device_driver *driver;
5633 const struct device *parent;
5635 if (len <= 0 || !buffer)
5639 parent = dev->dev.parent;
5644 driver = parent->driver;
5645 if (driver && driver->name)
5646 strlcpy(buffer, driver->name, len);
5650 static void __net_exit netdev_exit(struct net *net)
5652 kfree(net->dev_name_head);
5653 kfree(net->dev_index_head);
5656 static struct pernet_operations __net_initdata netdev_net_ops = {
5657 .init = netdev_init,
5658 .exit = netdev_exit,
5661 static void __net_exit default_device_exit(struct net *net)
5663 struct net_device *dev, *aux;
5665 * Push all migratable network devices back to the
5666 * initial network namespace
5669 for_each_netdev_safe(net, dev, aux) {
5671 char fb_name[IFNAMSIZ];
5673 /* Ignore unmoveable devices (i.e. loopback) */
5674 if (dev->features & NETIF_F_NETNS_LOCAL)
5677 /* Leave virtual devices for the generic cleanup */
5678 if (dev->rtnl_link_ops)
5681 /* Push remaing network devices to init_net */
5682 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5683 err = dev_change_net_namespace(dev, &init_net, fb_name);
5685 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5686 __func__, dev->name, err);
5693 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5695 /* At exit all network devices most be removed from a network
5696 * namespace. Do this in the reverse order of registeration.
5697 * Do this across as many network namespaces as possible to
5698 * improve batching efficiency.
5700 struct net_device *dev;
5702 LIST_HEAD(dev_kill_list);
5705 list_for_each_entry(net, net_list, exit_list) {
5706 for_each_netdev_reverse(net, dev) {
5707 if (dev->rtnl_link_ops)
5708 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5710 unregister_netdevice_queue(dev, &dev_kill_list);
5713 unregister_netdevice_many(&dev_kill_list);
5717 static struct pernet_operations __net_initdata default_device_ops = {
5718 .exit = default_device_exit,
5719 .exit_batch = default_device_exit_batch,
5723 * Initialize the DEV module. At boot time this walks the device list and
5724 * unhooks any devices that fail to initialise (normally hardware not
5725 * present) and leaves us with a valid list of present and active devices.
5730 * This is called single threaded during boot, so no need
5731 * to take the rtnl semaphore.
5733 static int __init net_dev_init(void)
5735 int i, rc = -ENOMEM;
5737 BUG_ON(!dev_boot_phase);
5739 if (dev_proc_init())
5742 if (netdev_kobject_init())
5745 INIT_LIST_HEAD(&ptype_all);
5746 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5747 INIT_LIST_HEAD(&ptype_base[i]);
5749 if (register_pernet_subsys(&netdev_net_ops))
5753 * Initialise the packet receive queues.
5756 for_each_possible_cpu(i) {
5757 struct softnet_data *queue;
5759 queue = &per_cpu(softnet_data, i);
5760 skb_queue_head_init(&queue->input_pkt_queue);
5761 queue->completion_queue = NULL;
5762 INIT_LIST_HEAD(&queue->poll_list);
5765 queue->csd.func = trigger_softirq;
5766 queue->csd.info = queue;
5767 queue->csd.flags = 0;
5770 queue->backlog.poll = process_backlog;
5771 queue->backlog.weight = weight_p;
5772 queue->backlog.gro_list = NULL;
5773 queue->backlog.gro_count = 0;
5778 /* The loopback device is special if any other network devices
5779 * is present in a network namespace the loopback device must
5780 * be present. Since we now dynamically allocate and free the
5781 * loopback device ensure this invariant is maintained by
5782 * keeping the loopback device as the first device on the
5783 * list of network devices. Ensuring the loopback devices
5784 * is the first device that appears and the last network device
5787 if (register_pernet_device(&loopback_net_ops))
5790 if (register_pernet_device(&default_device_ops))
5793 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5794 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5796 hotcpu_notifier(dev_cpu_callback, 0);
5804 subsys_initcall(net_dev_init);
5806 static int __init initialize_hashrnd(void)
5808 get_random_bytes(&hashrnd, sizeof(hashrnd));
5812 late_initcall_sync(initialize_hashrnd);